检索结果-内蒙古大学图书馆

On the Robustness of Adversarial Training Against Uncertainty Attacks

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Ledda, Emanuele Scodeller, Giovanni Angioni, Daniele Piras, Giorgio Cinà, Antonio Emanuele Fumera, Giorgio Biggio, Battista Roli, Fabio Department of Computer Control and Management Engineering Sapienza University of Rome Via Ariosto 25 Rome00185 Italy Department of Informatics Bioengineering Robotics and Systems Engineering University of Genova Via Dodecaneso 35 Genova16146 Italy Department of Electric and Electronic Engineering University of Cagliari Via Marengo 3 Cagliari09100 Italy

In learning problems, the noise inherent to the task at hand hinders the possibility to infer without a certain degree of uncertainty. Quantifying this uncertainty, regardless of its wide use, assumes high relevance for security-sensitive applications. Within these scenarios, it becomes fundamental to guarantee good (i.e., trustworthy) uncertainty measures, which downstream modules can securely employ to drive the final decision-making process. However, an attacker may be interested in forcing the system to produce either (i) highly uncertain outputs jeopardizing the system’s availability or (ii) low uncertainty estimates, making the system accept uncertain samples that would instead require a careful inspection (e.g., human intervention). Therefore, it becomes fundamental to understand how to obtain robust uncertainty estimates against these kinds of attacks. In this work, we reveal both empirically and theoretically that defending against adversarial examples, i.e., carefully perturbed samples that cause misclassification, additionally guarantees a more secure, trustworthy uncertainty estimate under common attack scenarios without the need for an ad-hoc defense strategy. To support our claims, we evaluate multiple adversarial-robust models from the publicly available benchmark RobustBench on the CIFAR-10 and ImageNet datasets. © 2024, CC BY.

关键词： Adversarial machine learning

Risk-Sensitive RL Using Sampling-Based Expectation-Maximization

学校读者我要写书评

暂无评论

Risk-Sensitive RL Using Sampling-Based Expectation-Maximizat...

IEEE Conference on Decision and control

作者： Erfaun Noorani John S. Baras Karl H. Johansson Department of Electrical and Computer Engineering Institute for Systems Research (ISR) at the University of Maryland College Park MD USA Automatic Control Lab Royal Institute of Technology Stockholm Sweden

There is a need for robust Reinforcement Learning (RL) algorithms that can cope with model misspecification, parameter uncertainty, disturbances, etc. Risk-sensitive methods offer an approach to developing robust RL algorithms by hedging against undesirable outcomes in a probabilistic manner. The Probabilistic Graphical Model (PGM) framework offers systematic exploration for risk-sensitive RL. In this paper, we bridge the Markov Decision Process (MDP) and the PGM frameworks. We exploit the equivalence of optimizing a certain risk-sensitive criterion in the MDP formalism with optimizing a log-likelihood objective in the PGM formalism. By utilizing this equivalence, we offer an approach for developing risk-sensitive algorithms by leveraging the PGM framework. We explore the Expectation-Maximization (EM) algorithm under the PGM formalism. We show that risk-sensitive policy gradient methods can be obtained by applying sampling-based approaches to the EM algorithm, e.g., Monte-Carlo EM, with the log-likelihood. We show that Monte-Carlo EM leads to a risk-sensitive Monte-Carlo policy gradient algorithm. Our simulations illustrate the risk-sensitive nature of the resulting algorithm.

关键词：

Diagnosing and Augmenting Feature Representations in Correctional Inverse Reinforcement Learning

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Lourenço, Inês Bobu, Andreea Rojas, Cristian R. Wahlberg, Bo The Division of Decision and Control Systems Kth Royal Institute of Technology Stockholm Sweden The Department of Electrical Engineering and Computer Sciences University of California Berkeley BerkeleyCA United States

Robots have been increasingly better at doing tasks for humans by learning from their feedback, but still often suffer from model misalignment due to missing or incorrectly learned features. When the features the robot needs to learn to perform its task are missing or do not generalize well to new settings, the robot will not be able to learn the task the human wants and, even worse, may learn a completely different and undesired behavior. Prior work shows how the robot can detect when its representation is missing some feature and can, thus, ask the human to be taught about the new feature;however, these works do not differentiate between features that are completely missing and those that exist but do not generalize to new environments. In the latter case, the robot would detect misalignment and simply learn a new feature, leading to an arbitrarily growing feature representation that can, in turn, lead to spurious correlations and incorrect learning down the line. In this work, we propose separating the two sources of misalignment: we propose a framework for determining whether a feature the robot needs is incorrectly learned and does not generalize to new environment setups vs. is entirely missing from the robot's representation. Once we detect the source of error, we show how the human can initiate the realignment process for the model: if the feature is missing, we follow prior work for learning new features;however, if the feature exists but does not generalize, we use data augmentation to expand its training and, thus, complete the correction. We demonstrate the proposed approach in experiments with a simulated 7DoF robot manipulator and physical human corrections. Copyright © 2023, The Authors. All rights reserved.

关键词： Manipulators

Vehicle type recognition based on audio data 58

学校读者我要写书评

暂无评论

Vehicle type recognition based on audio data

58th Hawaii International Conference on System Sciences, HICSS 2025

作者： Kobiela, Dariusz Hajdasz, Michal Erezman, Mateusz Nurzyńska, Karolina Zaporowski, Szymon Kurowski, Adam Weichbroth, Pawel Department of Software Engineering Faculty of Electronics Telecommunication and Informatics Gdańsk University of Technology Gabriela Narutowicza 11/12 Gdańsk80-233 Poland Department of Multimedia Systems Faculty of Electronics Telecommunication and Informatics Gdańsk University of Technology Gabriela Narutowicza 11/12 Gdańsk80-233 Poland Department of Algorithmics and Software Faculty of Automatic Control Electronics and Computer Science Silesian University of Technology Akademicka 16 Gliwice44-100 Poland

ISBN: (纸本)9780998133188

Identifying different vehicle types can help manage traffic more efficiently, reduce congestion, and improve public safety. This study aims to create a classification model that can recognize vehicle types based on the sound of passing vehicles. To achieve this, a database of raw audio files containing 1763 samples from two sources was assembled. The time-domain signals were converted to a time-frequency representation using the short-time Fourier transform to generate Mel Spectrograms. Mel-frequency Cepstral Coefficients (MFCCs) were also generated using the discrete cosine transform. In our experiments we compared these approaches. Since the data was imbalanced we applied online augmentation. Based on the literature review, we chose a Convolutional Neural Network (CNN) classifier because it is particularly well suited for analyzing large datasets due to its automatic feature extraction, parameter sharing and sparsity. The results showed that Mel Spectrograms were more effective for audio data preprocessing in this particular use case, achieving the highest accuracy of 0.875 and the highest f1-score of 0.877 compared to MFCCs. © 2025 IEEE computer Society. All rights reserved.

关键词： Image coding

Scalable and low-cost remote lab platforms: Teaching industrial robotics using open-source tools and understanding its social implications

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Kumar, Amit Jose, Jaison Jain, Archit Kulkarni, Siddharth Arya, Kavi Centre for Systems and Control Indian Institute of Technology Bombay Mumbai India Embedded Real-Time Systems / e-Yantra Lab. Indian Institute of Technology Bombay Mumbai India Computer Science and Engineering Department Indian Institute of Technology Bombay Mumbai India

With recent advancements in industrial robots, educating students in new technologies and preparing them for the future is imperative. However, access to industrial robots for teaching poses challenges, such as the high cost of acquiring these robots, the safety of the operator and the robot, and complicated training material. This paper proposes two low-cost platforms built using open-source tools like Robot Operating System (ROS) and its latest version ROS 2 to help students learn and test algorithms on remotely connected industrial robots. Universal Robotics (UR5) arm and a custom mobile rover were deployed in different life-size testbeds, a greenhouse, and a warehouse to create an Autonomous Agricultural Harvester System (AAHS) and an Autonomous Warehouse Management System (AWMS). These platforms were deployed for a period of 7 months and were tested for their efficacy with 1,433 and 1,312 students, respectively. The hardware used in AAHS and AWMS was controlled remotely for 160 and 355 hours, respectively, by students over a period of 3 months. © 2024, CC BY.

关键词： Costs

Differentially Private Dual Gradient Tracking for Distributed Resource Allocation

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Huo, Wei Chen, Xiaomeng Huang, Lingying Johansson, Karl Henrik Shi, Ling Department of Electronic and Computer Engineering Hong Kong University of Science and Technology Hong Kong School of Electrical and Electronic Engineering Nanyang Technology University Singapore639798 Singapore Division of Decision and Control Systems KTH Royal Institute of Technology Stockholm Sweden

This paper investigates privacy issues in distributed resource allocation over directed networks, where each agent holds a private cost function and optimizes its decision subject to a global coupling constraint through local interaction with other agents. Conventional methods for resource allocation over directed networks require all agents to transmit their original data to neighbors, which poses the risk of disclosing sensitive and private information. To address this issue, we propose an algorithm called differentially private dual gradient tracking (DP-DGT) for distributed resource allocation, which obfuscates the exchanged messages using independent Laplacian noise. Our algorithm ensures that the agents’ decisions converge to a neighborhood of the optimal solution almost surely. Furthermore, without the assumption of bounded gradients, we prove that the cumulative differential privacy loss under the proposed algorithm is finite even when the number of iterations goes to infinity. To the best of our knowledge, we are the first to simultaneously achieve these two goals in distributed resource allocation problems over directed networks. Finally, numerical simulations on economic dispatch problems within the IEEE 14-bus system illustrate the effectiveness of our proposed algorithm. © 2024, CC BY.

关键词： Differential privacy

Adaptive Sampling of Algal Blooms Using Autonomous Underwater Vehicle and Satellite Imagery: Experimental Validation in the Baltic Sea

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Fonseca, Joana Bhat, Sriharsha Lock, Matthew Stenius, Ivan Johansson, Karl H. The Division of Decision and Control Systems School of Electrical Engineering and Computer Science and Digital Futures Sweden The Department of Aeronautical and Vehicle Engineering School of Engineering Sciences Kth Royal Institute of Technology StockholmSE-100 44 Sweden

This paper investigates using satellite data to improve adaptive sampling missions, particularly for front tracking scenarios such as with algal blooms. Our proposed solution to find and track algal bloom fronts uses an Autonomous Underwater Vehicle (AUV) equipped with a sensor that measures the concentration of chlorophyll a and satellite data. The proposed method learns the kernel parameters for a Gaussian process (GP) model using satellite images of chlorophyll a from the previous days. Then, using the data collected by the AUV, it models chlorophyll a concentration online. We take the gradient of this model to obtain the direction of the algal bloom front and feed it to our control algorithm. The performance of this method is evaluated through realistic simulations for an algal bloom front in the Baltic sea, using the models of the AUV and the chlorophyll a sensor. We compare the performance of different estimation methods, from GP to curve interpolation using least squares. Sensitivity analysis is performed to evaluate the impact of sensor noise on the methods' performance. We implement our method on an AUV and run experiments in the Stockholm archipelago in the summer of 2022. Copyright © 2023, The Authors. All rights reserved.

关键词： Autonomous underwater vehicles

Probabilistic Tube-based control Synthesis of Stochastic Multi-Agent systems under Signal Temporal Logic

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Vlahakis, Eleftherios E. Lindemann, Lars Sopasakis, Pantelis Dimarogonas, Dimos V. Division of Decision and Control Systems School of Electrical Engineering and Computer Science KTH Royal Institute of Technology Stockholm10044 Sweden Thomas Lord Department of Computer Science Viterbi School of Engineering University of Southern California Los AngelesCA90089 United States School of Electronics Electrical Engineering and Computer Science Queen’s University Northern Ireland Belfast United Kingdom

We consider the control design of stochastic discrete-time linear multi-agent systems (MASs) under a global signal temporal logic (STL) specification to be satisfied at a predefined probability. By decomposing the dynamics into deterministic and error components, we construct a probabilistic reachable tube (PRT) as the Cartesian product of reachable sets of the individual error systems driven by disturbances lying in confidence regions (CRs) with a fixed probability. By bounding the PRT probability with the specification probability, we tighten all state constraints induced by the STL specification by solving tractable optimization problems over segments of the PRT, and relax the underlying stochastic problem with a deterministic one. This approach reduces conservatism compared to tightening guided by the STL structure. Additionally, we propose a recursively feasible algorithm to attack the resulting problem by decomposing it into agent-level subproblems, which are solved iteratively according to a scheduling policy. We demonstrate our method on a ten-agent system, where existing approaches are impractical. © 2024, CC BY.

关键词： Stochastic systems

Improving finite-difference time-domain memory bandwidth by using block floating-point arithmetic

学校读者我要写书评

暂无评论

Improving finite-difference time-domain memory bandwidth by ...

Telecommunications Forum (TELFOR)

作者： Stefan Pijetlovic Ivan Kastelan Milos Subotic Nebojsa Pjevalica RT-RK Institute for computer based systems Novi Sad Serbia Department of Computing and control engineering (Faculty of Technical Sciences) University of Novi Sad Novi Sad Serbia

Finite-difference time-domain is a numerical method used for modelling of computational electrodynamics. The method is resource intensive, especially regarding memory usage since multiple memory accesses are required per single computation. In other words, memory, or more precisely its speed, is a limiting factor for the overall performance. Existing solutions focus either on structures with great parallelism such as graphical processing units or custom designs. The former have issues with underutilized resources, while the latter require significant time and effort to achieve similar or slightly better performances. The approach taken during this research was to target the memory itself and find a way to improve the efficiency of memory usage across multiple platforms. The main contribution of the paper is using block floating-point in a novel way regarding the finite-difference time-domain method. The presented solution is a verification of earlier research, proving that the performance of this algorithm can indeed be improved using the suggested methods.

关键词：